The idea that software has those negative characteristics merely because it resides within a database, is laughable.
You could do something more sane like running lang-extensions on the db so you have access to a “real” language to specify your business logic in, but then you haven’t really changed anything except to run your app logic on the DB server and doesn’t really have anything to do with the RDBMS at that point.
Moving business logic into the schema/queries can be fine, but you still have the issues that you can’t document it particularly well, can’t version control it well, can’t test it well, can’t trust your performance tuning well, etc. Which all derives from DBMS expecting to be treated as “living systems”, and has nothing to do with knowing/not knowing SQL.
And the problem with modifying living systems is that you really don’t want to be fucking around with it because anything could potentially send it reeling (see: surgeons). You really want it to be stable… so moving your unstable business logic out is almost a necessity
moving business logic into schema/queries is usually fine, (and perhaps the only sane place for the DB to directly interact with business logic) though schema changes has all the issues described. Queries get to live with your application code, and so can be treated sanely, but you only really get to a good setup with a decent query builder library (ORM-haters smashing strings together like an ape is naturally full of issues, and ORM-lovers get all the fun of table/object model-mismatch and poor query generation). But then you run into the issue that every database library tries to be database-agnostic, and you lose access to all the fancy features your database provides
I've found that smashing strings together works excellently so long as:
- You use something like JavaScript's template strings so that interpolated fragments are inline
- You have named placeholder support (so that you're not trying to count parameter placements which gets hugely complex for optional parameters).
- You mainly use it for SELECTS and have minimal query builder for INSERTs/UPDATEs/UPSERTs
And you get full access to all of the database's features.
And it’s not just named parameters; with a sane query builder you don’t need to totally rewrite things when dealing with subqueries vs top-level, aggregations, etc.
As long as you don’t go so far as to introduce the horror of an ORM
This has actually been the road I've been heading down on my personal projects. If I were to use this for work I would use something like Kysely so I can have type safety and make sure the application layer is updated to support every migration, but I've taken time to really practice SQL and write queries & migrations manually and avoid ORMs completely. I love writing SQL by hand, however I know that doesn't scale without some automated checking going on.
What you will find though is any sufficiently complicated business application software implementing an ad hoc informally-specified bug-ridden slow implementation of half of what the RDBMS is giving you for free.
That’s exactly not my point. My point is that, knowing SQL, knowing what RDBMS’s have and what they are, RDBMS’s are a bad place to stuff your business logic. You have a variety of options that don’t work well. Schemas/queries are the only relevant place where RDBMS’s excel, and because it’s a living system, you don’t really want something so unstable as business logic to permeate through your schema significantly.
It’s not a matter of familiarity. It’s the wrong tool for the job being discussed.
Being able to roll forward and back with code and data (including an updated schema) that is synchronised and works is a huge win. In most places where I've worked with serious RDBMS systems developers have no access whatsoever to production. The fact that this isn't the case universally illustrates the strength of these RDBMS approaches.